For full text search please use the '?' prefix. e.g. ? Onboarding

AGENTS

Working in this research workspace

A Dendron vault of research notes. Read this before editing notes or drafting from them.

It sits inside the research vault so it travels with the notes through git and is discovered by VS Code, which treats each vault as a workspace folder.

What is here

Vaults are research (the main one), wiki, placecolour, megaflora and personal. Notes are flat markdown files named hierarchically with dots, so the prefix carries the type: concept.*, approach.*, teaching.*, precedent.*, agent.*, project.*, evidence.*, place.*, research.*. Respect that naming when adding notes.

These notes are drafts toward academic publication. Treat them as work in progress by a specialist, rather than as material to explain back to its author.

Manuscript folders are sometimes added to the workspace while a paper is in progress, then removed. Those are working drafts rather than notes. Leave them out of any index. Published versions arrive in the publication corpus instead.

Writing style

Three rules, and they matter because prose that signals machine authorship creates rework:

  • Active voice.
  • No em-dashes. Use a full stop or a comma instead.
  • No negative parallelism. Avoid "this is X, not Y" and "it's not about X, it's about Y". Write the claim positively: instead of "reachability matters, not size", write "reachability matters more than size".

These apply to HTML comments and image captions as well as body prose. Before saving an edit, check the file for — and for "not" following a comma.

Note structure

Imitate the existing pattern: an introductory paragraph, then ## Related Notes holding [[wiki-links]], then thematic sections often carrying comparison tables, then ## Implications for More-than-Human Design, then ## References holding Chicago 17th footnote definitions.

Every [[link]] must resolve to an existing note, and every [^n] must be both defined and cited. Check both before finishing.

Citations

Use Chicago 17th.

In these notes, write references out in full as prose. Do not introduce citation keys here. Quarto manuscripts use keys instead, and that difference is deliberate.

When a reference already appears anywhere in this workspace, copy that entry verbatim rather than reformatting it, including its short-DOI form such as https://doi.org/10/gmhmmj. The existing entries are maintained as correct, so reformatting introduces drift. Search before writing a bibliography:

grep -rn "AuthorSurname" --include="*.md" .

When proposing a reference that does not yet appear here, flag it explicitly so it can be verified and added to Zotero. In replies, separate references already held in Zotero from those needing acquisition.

Flag any detail you constructed rather than read from the source, such as a DOI pattern, a page range or a publisher. Report suspected errors in existing entries instead of silently fixing them.

The lab publication corpus

The Deep Design Lab's own publications are archived, converted and summarised in a SharePoint-synced folder, Deep Design Lab - Documents/Research/Publications. The drive letter differs per machine; on the machine where this was written it is under D:\The University of Melbourne\.

That folder's AGENTS.md is the authority on the corpus. Read it before answering any question about what the lab has published, argued or demonstrated. It covers the file layout, which layer answers which kind of question, what is deliberately excluded, and the rule that the PDF outranks the conversion. Do not rely on this file for any of that.

Search rather than recalling from memory:

python <Publications>/_tools/corpus_search.py "your query" --papers

These notes are indexed as a second body of text, so --source both searches publications and notes together and labels each result paper or note. After substantial writing, refresh the note index:

python <Publications>/_tools/notes_index.py
python <Publications>/_tools/corpus_search.py --source notes --build-index

Writing a Quarto manuscript

Manuscripts cite with Better BibTeX keys, and those keys are the corpus filenames. Seeing @roudavskiDesignAllLife2022 in a .qmd gives direct access to summaries/, markdown/ and pdf/ files of that exact name. Use that rather than searching. It is faster and removes any doubt about which paper is meant.

Zotero

Never modify the Zotero library. Read it through a copy, or through the local HTTP API at http://127.0.0.1:23119/api/ while Zotero is running. The tools find it automatically from Zotero's own preferences.

python <Publications>/_tools/zotero_query.py search "multispecies"

A new computer

Clone the vaults, let SharePoint sync the corpus, then run any tool once. Paths are detected and remembered in ~/.ddl-corpus.json, which stays on that machine. Confirm with:

python <Publications>/_tools/corpus_paths.py --show

The note index is derived data and does not sync. Rebuild it once per machine with the two commands above.